tailieunhanh - Business process improvement_13

Tham khảo tài liệu 'business process improvement_13', kỹ thuật - công nghệ, điện - điện tử phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | . Statistical significance Estimating sd effect The key in the above approach is to determine an estimate for sd effect . Three statistical approaches are common 1. Prior knowledge about Ỡ If a is known we can compute sd effect from the above expression and make use of a conservative normal-based 95 confidence interval by drawing the line at This method is rarely used in practice because ỈĨ is rarely known. 2. Replication in the experimental design Replication will allow ÍT to be estimated from the data without depending on the correctness of a deterministic model. This is a real benefit. On the other hand the downside of such replication is that it increases the number of runs time and expense of the experiment. If replication can be afforded this method should be used. In such a case the analyst separates important from unimportant terms by drawing the line at with t denoting the percent point from the appropriate Student s-t distribution. 3. Assume 3-factor interactions and higher are zero This approach assumes away all 3-factor interactions and higher and uses the data pertaining to these interactions to estimate m Specifically with h denoting the number of 3-factor interactions and higher and SSQ is the sum of squares for these higher-order effects. The analyst separates important from unimportant effects by drawing the line at t sdfeffttt t -7 x ĩl with t denoting the percent point from the appropriate http div898 handbook pri section5 2 of 3 5 1 2006 10 31 32 AM . Statistical significance with h degrees of freedom Student s-t distribution. This method warrants caution o it involves an untestable assumption that such interactions 0 o it can result in an estimate for sd effect based on few terms even a single term and o it is virtually unusable for highly-fractionated designs since high-order interactions are not directly estimable . Non-statistical considerations The above statistical methods can and .